多任务学习对预测新抗原-MHCII类结合性的影响
IEEE transactions on computational biology and bioinformatics
|August 14, 2025
概括
对治疗性癌症疫苗来说,预测癌症新表位组与主要基因相容性复合体 (MHC) II类结合至关重要. 一种新的深度学习模型提高了预测准确性,超过了现有的新抗原识别方法.
科学领域:
- 免疫学和计算生物学
- 癌症研究和疫苗开发
背景情况:
- 瘤新进位是癌症疫苗的关键标,刺激免疫反应对抗癌细胞.
- 有效的新型疫苗取决于对抗原与主要基因相容性复合体 (MHC) 分子的结合的准确预测.
- 目前用于预测表位体-MHC结合的计算模型受到偏差数据集和MHCII类等位基因的不良性能的限制.
研究的目的:
- 开发一种改进的计算方法,用于预测新表位粒-MHCII类结合亲缘关系.
- 解决现有模型的局限性,特别是它们对各种MHCII类等位基因的表现和有限的新表位基因数据的局限性.
主要方法:
- 开发了一种使用多任务双向长短期记忆 (Bi-LSTM) 网络的新型深度学习模型.
- 该模型使用MHC类I和II训练数据之间的参数共享来增强有限数据集的预测.
- 使用接收器操作特征曲线 (AUC-ROC) 下的面积来评估性能.
主要成果:
- 多任务深度学习模型显著改善了癌症新表皮质-MHCII类结合的预测性能.
- 该模型实现了82.2%的AUC-ROC,超过了最先进的单基因新抗原预测模型.
- 拟议的模型表现出强大的泛化性能,表明其在不同数据集中的稳定性.
结论:
- 多任务学习有效地提高了预测新位基-MHCII类结合亲和关系,即使训练数据有限.
- 这种新的深度学习方法为识别癌症疫苗新位候选人提供了更准确,更可靠的工具.
- 改进的预测能力对推进个性化癌症免疫疗法的发展具有重大前景.
相关概念视频
Antigens Involved in Adaptive Immunity
669
An antigen is any substance the immune system identifies as foreign and potentially harmful to the body, prompting an immune response. Antigens have two functional properties: immunogenicity and reactivity. Immunogenicity is the ability of an antigen to stimulate a specific immune response. At the same time, reactivity describes the antigen's ability to react with the cells and antibodies produced in response to it.
Complete Antigens
Complete antigens possess both immunogenicity and...
Complete Antigens
Complete antigens possess both immunogenicity and...
669
T Cell Activation and Clonal Selection
4.6K
T cells are integral to our adaptive immune system, recognizing and effectively responding to foreign antigens. T cell activation and clonal selection are pivotal in orchestrating this immune response. This article elucidates these mechanisms, detailing the roles of cluster of differentiation (CD) markers, major histocompatibility complex (MHC) molecules, costimulatory signals, and the process of clonal selection.
Naive T cells that have not yet encountered an antigen express two primary CD...
Naive T cells that have not yet encountered an antigen express two primary CD...
4.6K


